diff --git a/opencl/source/context/context.cpp b/opencl/source/context/context.cpp
index f7db6d64a7..6021d02b23 100644
--- a/opencl/source/context/context.cpp
+++ b/opencl/source/context/context.cpp
@@ -158,6 +158,21 @@ uint32_t Context::getMaxRootDeviceIndex() const {
 }
 
 CommandQueue *Context::getSpecialQueue(uint32_t rootDeviceIndex) {
+    if (specialQueues[rootDeviceIndex])
+        return specialQueues[rootDeviceIndex];
+
+    static std::mutex mtx;
+    std::lock_guard lock(mtx);
+
+    if (!specialQueues[rootDeviceIndex]) {
+        cl_int errcodeRet = CL_SUCCESS;
+        auto deviceOrdinal = std::distance(this->getRootDeviceIndices().begin(), std::find(this->getRootDeviceIndices().begin(), this->getRootDeviceIndices().end(), rootDeviceIndex));
+        auto commandQueue = CommandQueue::create(this, this->getDevice(deviceOrdinal), nullptr, true, errcodeRet);
+        DEBUG_BREAK_IF(commandQueue == nullptr);
+        DEBUG_BREAK_IF(errcodeRet != CL_SUCCESS);
+        overrideSpecialQueueAndDecrementRefCount(commandQueue, rootDeviceIndex);
+    }
+
     return specialQueues[rootDeviceIndex];
 }
 
@@ -179,6 +194,7 @@ bool Context::createImpl(const cl_context_properties *properties,
                          const ClDeviceVector &inputDevices,
                          void(CL_CALLBACK *funcNotify)(const char *, const void *, size_t, void *),
                          void *data, cl_int &errcodeRet) {
+    errcodeRet = CL_SUCCESS;
 
     auto propertiesCurrent = properties;
     bool interopUserSync = false;
@@ -288,14 +304,6 @@ bool Context::createImpl(const cl_context_properties *properties,
         }
     }
 
-    for (auto &device : devices) {
-        if (!specialQueues[device->getRootDeviceIndex()]) {
-            auto commandQueue = CommandQueue::create(this, device, nullptr, true, errcodeRet); // NOLINT(clang-analyzer-cplusplus.NewDelete)
-            DEBUG_BREAK_IF(commandQueue == nullptr);
-            overrideSpecialQueueAndDecrementRefCount(commandQueue, device->getRootDeviceIndex());
-        }
-    }
-
     return true;
 }
 
diff --git a/opencl/test/unit_test/context/context_tests.cpp b/opencl/test/unit_test/context/context_tests.cpp
index 98d4f20e74..4ddda11e6b 100644
--- a/opencl/test/unit_test/context/context_tests.cpp
+++ b/opencl/test/unit_test/context/context_tests.cpp
@@ -120,7 +120,7 @@ TEST_F(ContextTest, WhenCreatingContextThenSpecialQueueIsAvailable) {
 TEST_F(ContextTest, WhenSettingSpecialQueueThenQueueIsAvailable) {
     MockContext context((ClDevice *)devices[0], true);
 
-    auto specialQ = context.getSpecialQueue(0u);
+    auto specialQ = context.specialQueues[0];
     EXPECT_EQ(specialQ, nullptr);
 
     auto cmdQ = new MockCommandQueue(&context, (ClDevice *)devices[0], 0, false);
diff --git a/opencl/test/unit_test/gtpin/gtpin_tests.cpp b/opencl/test/unit_test/gtpin/gtpin_tests.cpp
index 8cf0bbc86e..28b00a591e 100644
--- a/opencl/test/unit_test/gtpin/gtpin_tests.cpp
+++ b/opencl/test/unit_test/gtpin/gtpin_tests.cpp
@@ -435,7 +435,7 @@ TEST_F(GTPinTests, givenInvalidArgumentsThenBufferUnMapFails) {
 }
 
 TEST_F(GTPinTests, givenValidRequestForHugeMemoryAllocationThenBufferAllocateFails) {
-
+    [[maybe_unused]] auto cmdQ = pContext->getSpecialQueue(0);
     DebugManagerStateRestore restorer;
     for (auto &allocationInUSMShared : ::testing::Bool()) {
         debugManager.flags.GTPinAllocateBufferInSharedMemory.set(allocationInUSMShared);