diff --git a/sycl/source/detail/queue_impl.cpp b/sycl/source/detail/queue_impl.cpp
index 24ed44b219e3a..27b62c0118ee8 100644
--- a/sycl/source/detail/queue_impl.cpp
+++ b/sycl/source/detail/queue_impl.cpp
@@ -859,6 +859,13 @@ void queue_impl::verifyProps(const property_list &Props) const {
                                                 CheckPropertiesWithData);
 }
 
+std::vector<ur_exp_kernel_arg_properties_t> &
+queue_impl::getKernelArgStorage(uint32_t size) {
+  MKernelArgStorage.clear();
+  MKernelArgStorage.reserve(size);
+  return MKernelArgStorage;
+}
+
 } // namespace detail
 } // namespace _V1
 } // namespace sycl
diff --git a/sycl/source/detail/queue_impl.hpp b/sycl/source/detail/queue_impl.hpp
index 7e15c772f2697..f61bb23df563d 100644
--- a/sycl/source/detail/queue_impl.hpp
+++ b/sycl/source/detail/queue_impl.hpp
@@ -683,6 +683,11 @@ class queue_impl : public std::enable_shared_from_this<queue_impl> {
   }
 #endif
 
+  /// Clears MKernelArgsStorage, has it .reserve(size), and returns a reference
+  /// to it. Not inherently thread safe.
+  std::vector<ur_exp_kernel_arg_properties_t> &
+  getKernelArgStorage(uint32_t size);
+
 protected:
   template <typename HandlerType = handler>
   EventImplPtr insertHelperBarrier(const HandlerType &Handler) {
@@ -999,6 +1004,10 @@ class queue_impl : public std::enable_shared_from_this<queue_impl> {
 
   ur_queue_handle_t MQueue;
 
+  // To avoid re-allocating this every time a kernel is enqueued we keep this
+  // vector around and .clear()/.reserve() for each kernel instead.
+  std::vector<ur_exp_kernel_arg_properties_t> MKernelArgStorage;
+
   // Access should be guarded with MMutex
   struct DependencyTrackingItems {
     // This event is employed for enhanced dependency tracking with in-order
diff --git a/sycl/source/detail/scheduler/commands.cpp b/sycl/source/detail/scheduler/commands.cpp
index 70f12d0a59ef7..94c704db22f1e 100644
--- a/sycl/source/detail/scheduler/commands.cpp
+++ b/sycl/source/detail/scheduler/commands.cpp
@@ -2424,8 +2424,7 @@ static ur_result_t SetKernelParamsAndLaunch(
         DeviceImageImpl ? DeviceImageImpl->get_spec_const_blob_ref() : Empty);
   }
 
-  std::vector<ur_exp_kernel_arg_properties_t> UrArgs;
-  UrArgs.reserve(Args.size());
+  auto UrArgs = Queue.getKernelArgStorage(Args.size());
 
   if (KernelFuncPtr && !KernelHasSpecialCaptures) {
     auto setFunc = [&UrArgs,