Accelerating Performance of GPU-based Workloads Using CXL