diff --git a/Makefile b/Makefile
index 9ab37ac..9894c5c 100644
--- a/Makefile
+++ b/Makefile
@@ -189,6 +189,7 @@
 				-Iinclude/lib/cpus/${ARCH}		\
 				-Iinclude/lib/el3_runtime		\
 				-Iinclude/lib/el3_runtime/${ARCH}	\
+				-Iinclude/lib/extensions		\
 				-Iinclude/lib/pmf			\
 				-Iinclude/lib/psci			\
 				-Iinclude/lib/xlat_tables		\
diff --git a/bl31/bl31.mk b/bl31/bl31.mk
index 336c295..fccdc8a 100644
--- a/bl31/bl31.mk
+++ b/bl31/bl31.mk
@@ -46,6 +46,10 @@
 				services/std_svc/sdei/sdei_state.c
 endif
 
+ifeq (${ENABLE_SPE_FOR_LOWER_ELS},1)
+BL31_SOURCES		+=	lib/extensions/spe/spe.c
+endif
+
 BL31_LINKERFILE		:=	bl31/bl31.ld.S
 
 # Flag used to indicate if Crash reporting via console should be included
diff --git a/docs/user-guide.rst b/docs/user-guide.rst
index 172e793..eeacc16 100644
--- a/docs/user-guide.rst
+++ b/docs/user-guide.rst
@@ -344,9 +344,9 @@
    the ``ENABLE_PMF`` build option as well. Default is 0.
 
 -  ``ENABLE_SPE_FOR_LOWER_ELS`` : Boolean option to enable Statistical Profiling
-   extensions. This is an optional architectural feature available only for
-   AArch64 8.2 onwards. This option defaults to 1 but is automatically
-   disabled when the target architecture is AArch32 or AArch64 8.0/8.1.
+   extensions. This is an optional architectural feature for AArch64.
+   The default is 1 but is automatically disabled when the target architecture
+   is AArch32.
 
 -  ``ENABLE_STACK_PROTECTOR``: String option to enable the stack protection
    checks in GCC. Allowed values are "all", "strong" and "0" (default).
diff --git a/include/common/aarch64/el3_common_macros.S b/include/common/aarch64/el3_common_macros.S
index 34fdaee..ed35df8 100644
--- a/include/common/aarch64/el3_common_macros.S
+++ b/include/common/aarch64/el3_common_macros.S
@@ -95,10 +95,6 @@
 	 * MDCR_EL3.SPD32: Set to 0b10 to disable AArch32 Secure self-hosted
 	 *  privileged debug from S-EL1.
 	 *
-	 * MDCR_EL3.NSPB (ARM v8.2): SPE enabled in non-secure state and
-	 * disabled in secure state. Accesses to SPE registers at SEL1 generate
-	 * trap exceptions to EL3.
-	 *
 	 * MDCR_EL3.TDOSA: Set to zero so that EL2 and EL2 System register
 	 *  access to the powerdown debug registers do not trap to EL3.
 	 *
@@ -112,19 +108,6 @@
 	 */
 	mov_imm	x0, ((MDCR_EL3_RESET_VAL | MDCR_SDD_BIT | MDCR_SPD32(MDCR_SPD32_DISABLE)) \
 			& ~(MDCR_TDOSA_BIT | MDCR_TDA_BIT | MDCR_TPM_BIT))
-
-#if ENABLE_SPE_FOR_LOWER_ELS
-	/* Detect if SPE is implemented */
-	mrs	x1, id_aa64dfr0_el1
-	ubfx	x1, x1, #ID_AA64DFR0_PMS_SHIFT, #ID_AA64DFR0_PMS_LENGTH
-	cmp	x1, #0x1
-	b.ne	1f
-
-	/* Enable SPE for use by normal world */
-	orr	x0, x0, #MDCR_NSPB(MDCR_NSPB_EL1)
-1:
-#endif
-
 	msr	mdcr_el3, x0
 
 	/* ---------------------------------------------------------------------
diff --git a/include/lib/aarch64/arch.h b/include/lib/aarch64/arch.h
index cb7dab7..777a01a 100644
--- a/include/lib/aarch64/arch.h
+++ b/include/lib/aarch64/arch.h
@@ -604,4 +604,9 @@
 #define PAR_ADDR_SHIFT	12
 #define PAR_ADDR_MASK	(BIT(40) - 1) /* 40-bits-wide page address */
 
+/*******************************************************************************
+ * Definitions for system register interface to SPE
+ ******************************************************************************/
+#define PMBLIMITR_EL1		S3_0_C9_C10_0
+
 #endif /* __ARCH_H__ */
diff --git a/include/lib/aarch64/arch_helpers.h b/include/lib/aarch64/arch_helpers.h
index 782343d..46d9a1c 100644
--- a/include/lib/aarch64/arch_helpers.h
+++ b/include/lib/aarch64/arch_helpers.h
@@ -197,6 +197,7 @@
 DEFINE_SYSOP_TYPE_FUNC(dmb, st)
 DEFINE_SYSOP_TYPE_FUNC(dmb, ld)
 DEFINE_SYSOP_TYPE_FUNC(dsb, ish)
+DEFINE_SYSOP_TYPE_FUNC(dsb, nsh)
 DEFINE_SYSOP_TYPE_FUNC(dsb, ishst)
 DEFINE_SYSOP_TYPE_FUNC(dmb, ish)
 DEFINE_SYSOP_TYPE_FUNC(dmb, ishst)
@@ -301,6 +302,7 @@
 DEFINE_SYSREG_READ_FUNC(ctr_el0)
 
 DEFINE_SYSREG_RW_FUNCS(mdcr_el2)
+DEFINE_SYSREG_RW_FUNCS(mdcr_el3)
 DEFINE_SYSREG_RW_FUNCS(hstr_el2)
 DEFINE_SYSREG_RW_FUNCS(cnthp_ctl_el2)
 DEFINE_SYSREG_RW_FUNCS(pmcr_el0)
@@ -320,6 +322,7 @@
 DEFINE_RENAME_SYSREG_WRITE_FUNC(icc_eoir1_el1, ICC_EOIR1_EL1)
 DEFINE_RENAME_SYSREG_WRITE_FUNC(icc_sgi0r_el1, ICC_SGI0R_EL1)
 
+DEFINE_RENAME_SYSREG_RW_FUNCS(pmblimitr_el1, PMBLIMITR_EL1)
 
 #define IS_IN_EL(x) \
 	(GET_EL(read_CurrentEl()) == MODE_EL##x)
diff --git a/include/lib/el3_runtime/aarch64/context.h b/include/lib/el3_runtime/aarch64/context.h
index cf06a64..5889904 100644
--- a/include/lib/el3_runtime/aarch64/context.h
+++ b/include/lib/el3_runtime/aarch64/context.h
@@ -313,7 +313,6 @@
  * Function prototypes
  ******************************************************************************/
 void el1_sysregs_context_save(el1_sys_regs_t *regs);
-void el1_sysregs_context_save_post_ops(void);
 void el1_sysregs_context_restore(el1_sys_regs_t *regs);
 #if CTX_INCLUDE_FPREGS
 void fpregs_context_save(fp_regs_t *regs);
diff --git a/include/lib/extensions/spe.h b/include/lib/extensions/spe.h
new file mode 100644
index 0000000..8a74127
--- /dev/null
+++ b/include/lib/extensions/spe.h
@@ -0,0 +1,13 @@
+/*
+ * Copyright (c) 2017, ARM Limited and Contributors. All rights reserved.
+ *
+ * SPDX-License-Identifier: BSD-3-Clause
+ */
+
+#ifndef __SPE_H__
+#define __SPE_H__
+
+void spe_enable(int el2_unused);
+void spe_disable(void);
+
+#endif /* __SPE_H__ */
diff --git a/include/plat/arm/common/plat_arm.h b/include/plat/arm/common/plat_arm.h
index a28a903..abd7395 100644
--- a/include/plat/arm/common/plat_arm.h
+++ b/include/plat/arm/common/plat_arm.h
@@ -227,7 +227,4 @@
 		uint32_t cookie_lo,
 		void *handle);
 
-/* Disable Statistical Profiling Extensions helper */
-void arm_disable_spe(void);
-
 #endif /* __PLAT_ARM_H__ */
diff --git a/lib/el3_runtime/aarch32/context_mgmt.c b/lib/el3_runtime/aarch32/context_mgmt.c
index 3e7a5b7..a8672d6 100644
--- a/lib/el3_runtime/aarch32/context_mgmt.c
+++ b/lib/el3_runtime/aarch32/context_mgmt.c
@@ -125,6 +125,17 @@
 }
 
 /*******************************************************************************
+ * Enable architecture extensions on first entry to Non-secure world.
+ * When EL2 is implemented but unused `el2_unused` is non-zero, otherwise
+ * it is zero.
+ ******************************************************************************/
+static void enable_extensions_nonsecure(int el2_unused)
+{
+#if IMAGE_BL32
+#endif
+}
+
+/*******************************************************************************
  * The following function initializes the cpu_context for a CPU specified by
  * its `cpu_idx` for first use, and sets the initial entrypoint state as
  * specified by the entry_point_info structure.
@@ -161,6 +172,7 @@
 {
 	uint32_t hsctlr, scr;
 	cpu_context_t *ctx = cm_get_context(security_state);
+	int el2_unused = 0;
 
 	assert(ctx);
 
@@ -185,6 +197,8 @@
 			isb();
 		} else if (read_id_pfr1() &
 			(ID_PFR1_VIRTEXT_MASK << ID_PFR1_VIRTEXT_SHIFT)) {
+			el2_unused = 1;
+
 			/*
 			 * Set the NS bit to access NS copies of certain banked
 			 * registers
@@ -283,5 +297,6 @@
 			write_scr(read_scr() & ~SCR_NS_BIT);
 			isb();
 		}
+		enable_extensions_nonsecure(el2_unused);
 	}
 }
diff --git a/lib/el3_runtime/aarch64/context.S b/lib/el3_runtime/aarch64/context.S
index 143da95..620ec16 100644
--- a/lib/el3_runtime/aarch64/context.S
+++ b/lib/el3_runtime/aarch64/context.S
@@ -9,7 +9,6 @@
 #include <context.h>
 
 	.global	el1_sysregs_context_save
-	.global el1_sysregs_context_save_post_ops
 	.global	el1_sysregs_context_restore
 #if CTX_INCLUDE_FPREGS
 	.global	fpregs_context_save
@@ -112,36 +111,6 @@
 /* -----------------------------------------------------
  * The following function strictly follows the AArch64
  * PCS to use x9-x17 (temporary caller-saved registers)
- * to do post operations after saving the EL1 system
- * register context.
- * -----------------------------------------------------
- */
-func el1_sysregs_context_save_post_ops
-#if ENABLE_SPE_FOR_LOWER_ELS
-	/* Detect if SPE is implemented */
-	mrs	x9, id_aa64dfr0_el1
-	ubfx	x9, x9, #ID_AA64DFR0_PMS_SHIFT, #ID_AA64DFR0_PMS_LENGTH
-	cmp	x9, #0x1
-	b.ne	1f
-
-	/*
-	 * Before switching from normal world to secure world
-	 * the profiling buffers need to be drained out to memory.  This is
-	 * required to avoid an invalid memory access when TTBR is switched
-	 * for entry to SEL1.
-	 */
-	.arch	armv8.2-a+profile
-	psb	csync
-	dsb	nsh
-	.arch	armv8-a
-1:
-#endif
-	ret
-endfunc el1_sysregs_context_save_post_ops
-
-/* -----------------------------------------------------
- * The following function strictly follows the AArch64
- * PCS to use x9-x17 (temporary caller-saved registers)
  * to restore EL1 system register context.  It assumes
  * that 'x0' is pointing to a 'el1_sys_regs' structure
  * from where the register context will be restored
diff --git a/lib/el3_runtime/aarch64/context_mgmt.c b/lib/el3_runtime/aarch64/context_mgmt.c
index c8232df..8f1523f 100644
--- a/lib/el3_runtime/aarch64/context_mgmt.c
+++ b/lib/el3_runtime/aarch64/context_mgmt.c
@@ -15,6 +15,7 @@
 #include <platform_def.h>
 #include <pubsub_events.h>
 #include <smcc_helpers.h>
+#include <spe.h>
 #include <string.h>
 #include <utils.h>
 
@@ -209,6 +210,20 @@
 }
 
 /*******************************************************************************
+ * Enable architecture extensions on first entry to Non-secure world.
+ * When EL2 is implemented but unused `el2_unused` is non-zero, otherwise
+ * it is zero.
+ ******************************************************************************/
+static void enable_extensions_nonsecure(int el2_unused)
+{
+#if IMAGE_BL31
+#if ENABLE_SPE_FOR_LOWER_ELS
+	spe_enable(el2_unused);
+#endif
+#endif
+}
+
+/*******************************************************************************
  * The following function initializes the cpu_context for a CPU specified by
  * its `cpu_idx` for first use, and sets the initial entrypoint state as
  * specified by the entry_point_info structure.
@@ -245,6 +260,7 @@
 {
 	uint32_t sctlr_elx, scr_el3, mdcr_el2;
 	cpu_context_t *ctx = cm_get_context(security_state);
+	int el2_unused = 0;
 
 	assert(ctx);
 
@@ -258,6 +274,8 @@
 			sctlr_elx |= SCTLR_EL2_RES1;
 			write_sctlr_el2(sctlr_elx);
 		} else if (EL_IMPLEMENTED(2)) {
+			el2_unused = 1;
+
 			/*
 			 * EL2 present but unused, need to disable safely.
 			 * SCTLR_EL2 can be ignored in this case.
@@ -340,13 +358,6 @@
 			 * relying on hw. Some fields are architecturally
 			 * UNKNOWN on reset.
 			 *
-			 * MDCR_EL2.TPMS (ARM v8.2): Do not trap statistical
-			 * profiling controls to EL2.
-			 *
-			 * MDCR_EL2.E2PB (ARM v8.2): SPE enabled in non-secure
-			 * state. Accesses to profiling buffer controls at
-			 * non-secure EL1 are not trapped to EL2.
-			 *
 			 * MDCR_EL2.TDRA: Set to zero so that Non-secure EL0 and
 			 *  EL1 System register accesses to the Debug ROM
 			 *  registers are not trapped to EL2.
@@ -383,22 +394,6 @@
 					| MDCR_EL2_HPME_BIT | MDCR_EL2_TPM_BIT
 					| MDCR_EL2_TPMCR_BIT));
 
-#if ENABLE_SPE_FOR_LOWER_ELS
-			uint64_t id_aa64dfr0_el1;
-
-			/* Detect if SPE is implemented */
-			id_aa64dfr0_el1 = read_id_aa64dfr0_el1() >>
-				ID_AA64DFR0_PMS_SHIFT;
-			if ((id_aa64dfr0_el1 & ID_AA64DFR0_PMS_MASK) == 1) {
-				/*
-				 * Make sure traps to EL2 are not generated if
-				 * EL2 is implemented but not used.
-				 */
-				mdcr_el2 &= ~MDCR_EL2_TPMS;
-				mdcr_el2 |= MDCR_EL2_E2PB(MDCR_EL2_E2PB_EL1);
-			}
-#endif
-
 			write_mdcr_el2(mdcr_el2);
 
 			/*
@@ -420,6 +415,7 @@
 			write_cnthp_ctl_el2(CNTHP_CTL_RESET_VAL &
 						~(CNTHP_CTL_ENABLE_BIT));
 		}
+		enable_extensions_nonsecure(el2_unused);
 	}
 
 	cm_el1_sysregs_context_restore(security_state);
@@ -439,7 +435,6 @@
 	assert(ctx);
 
 	el1_sysregs_context_save(get_sysregs_ctx(ctx));
-	el1_sysregs_context_save_post_ops();
 
 #if IMAGE_BL31
 	if (security_state == SECURE)
diff --git a/lib/extensions/spe/spe.c b/lib/extensions/spe/spe.c
new file mode 100644
index 0000000..3b297f2
--- /dev/null
+++ b/lib/extensions/spe/spe.c
@@ -0,0 +1,85 @@
+/*
+ * Copyright (c) 2017, ARM Limited and Contributors. All rights reserved.
+ *
+ * SPDX-License-Identifier: BSD-3-Clause
+ */
+
+#include <arch.h>
+#include <arch_helpers.h>
+#include <pubsub.h>
+
+/*
+ * The assembler does not yet understand the psb csync mnemonic
+ * so use the equivalent hint instruction.
+ */
+#define psb_csync()	asm volatile("hint #17")
+
+void spe_enable(int el2_unused)
+{
+	uint64_t features;
+
+	features = read_id_aa64dfr0_el1() >> ID_AA64DFR0_PMS_SHIFT;
+	if ((features & ID_AA64DFR0_PMS_MASK) == 1) {
+		uint64_t v;
+
+		if (el2_unused) {
+			/*
+			 * MDCR_EL2.TPMS (ARM v8.2): Do not trap statistical
+			 * profiling controls to EL2.
+			 *
+			 * MDCR_EL2.E2PB (ARM v8.2): SPE enabled in Non-secure
+			 * state. Accesses to profiling buffer controls at
+			 * Non-secure EL1 are not trapped to EL2.
+			 */
+			v = read_mdcr_el2();
+			v &= ~MDCR_EL2_TPMS;
+			v |= MDCR_EL2_E2PB(MDCR_EL2_E2PB_EL1);
+			write_mdcr_el2(v);
+		}
+
+		/*
+		 * MDCR_EL2.NSPB (ARM v8.2): SPE enabled in Non-secure state
+		 * and disabled in secure state. Accesses to SPE registers at
+		 * S-EL1 generate trap exceptions to EL3.
+		 */
+		v = read_mdcr_el3();
+		v |= MDCR_NSPB(MDCR_NSPB_EL1);
+		write_mdcr_el3(v);
+	}
+}
+
+void spe_disable(void)
+{
+	uint64_t features;
+
+	features = read_id_aa64dfr0_el1() >> ID_AA64DFR0_PMS_SHIFT;
+	if ((features & ID_AA64DFR0_PMS_MASK) == 1) {
+		uint64_t v;
+
+		/* Drain buffered data */
+		psb_csync();
+		dsbnsh();
+
+		/* Disable profiling buffer */
+		v = read_pmblimitr_el1();
+		v &= ~(1ULL << 0);
+		write_pmblimitr_el1(v);
+		isb();
+	}
+}
+
+static void *spe_drain_buffers_hook(const void *arg)
+{
+	uint64_t features;
+
+	features = read_id_aa64dfr0_el1() >> ID_AA64DFR0_PMS_SHIFT;
+	if ((features & ID_AA64DFR0_PMS_MASK) == 1) {
+		/* Drain buffered data */
+		psb_csync();
+		dsbnsh();
+	}
+
+	return 0;
+}
+
+SUBSCRIBE_TO_EVENT(cm_entering_secure_world, spe_drain_buffers_hook);
diff --git a/make_helpers/defaults.mk b/make_helpers/defaults.mk
index 660e54e..b7ce051 100644
--- a/make_helpers/defaults.mk
+++ b/make_helpers/defaults.mk
@@ -149,19 +149,10 @@
 # platforms).
 WARMBOOT_ENABLE_DCACHE_EARLY	:= 0
 
-# By default, enable Statistical Profiling Extensions.
-# The top level Makefile will disable this feature depending on
-# the target architecture and version number.
+# Build option to enable/disable the Statistical Profiling Extensions
 ENABLE_SPE_FOR_LOWER_ELS	:= 1
 
-# SPE is enabled by default but only supported on AArch64 8.2 onwards.
-# Disable it in all other cases.
+# SPE is only supported on AArch64 so disable it on AArch32.
 ifeq (${ARCH},aarch32)
     override ENABLE_SPE_FOR_LOWER_ELS := 0
-else
-    ifeq (${ARM_ARCH_MAJOR},8)
-        ifeq ($(ARM_ARCH_MINOR),$(filter $(ARM_ARCH_MINOR),0 1))
-            ENABLE_SPE_FOR_LOWER_ELS := 0
-        endif
-    endif
 endif
diff --git a/plat/arm/board/fvp/fvp_pm.c b/plat/arm/board/fvp/fvp_pm.c
index 0ab5b82..13bd8f2 100644
--- a/plat/arm/board/fvp/fvp_pm.c
+++ b/plat/arm/board/fvp/fvp_pm.c
@@ -14,6 +14,7 @@
 #include <plat_arm.h>
 #include <platform.h>
 #include <psci.h>
+#include <spe.h>
 #include <v2m_def.h>
 #include "drivers/pwrc/fvp_pwrc.h"
 #include "fvp_def.h"
@@ -57,7 +58,7 @@
 	 * On power down we need to disable statistical profiling extensions
 	 * before exiting coherency.
 	 */
-	arm_disable_spe();
+	spe_disable();
 #endif
 
 	/* Disable coherency if this cluster is to be turned off */
diff --git a/plat/arm/common/aarch64/arm_helpers.S b/plat/arm/common/aarch64/arm_helpers.S
index b53e60d..9d3a108 100644
--- a/plat/arm/common/aarch64/arm_helpers.S
+++ b/plat/arm/common/aarch64/arm_helpers.S
@@ -12,7 +12,6 @@
 	.globl	plat_crash_console_putc
 	.globl	plat_crash_console_flush
 	.globl	platform_mem_init
-	.globl	arm_disable_spe
 
 
 	/* -----------------------------------------------------
@@ -88,34 +87,6 @@
 	ret
 endfunc platform_mem_init
 
-	/* -----------------------------------------------------
-	 * void arm_disable_spe (void);
-	 * -----------------------------------------------------
-	 */
-#if ENABLE_SPE_FOR_LOWER_ELS
-func arm_disable_spe
-	/* Detect if SPE is implemented */
-	mrs	x0, id_aa64dfr0_el1
-	ubfx	x0, x0, #ID_AA64DFR0_PMS_SHIFT, #ID_AA64DFR0_PMS_LENGTH
-	cmp	x0, #0x1
-	b.ne	1f
-
-	/* Drain buffered data */
-	.arch	armv8.2-a+profile
-	psb	csync
-	dsb	nsh
-
-	/* Disable Profiling Buffer */
-	mrs	x0, pmblimitr_el1
-	bic	x0, x0, #1
-	msr	pmblimitr_el1, x0
-	isb
-	.arch	armv8-a
-1:
-	ret
-endfunc arm_disable_spe
-#endif
-
 /*
  * Need to use coherent stack when ARM Cryptocell is used to autheticate images
  * since Cryptocell uses DMA to transfer data and it is not coherent with the