Low-Latency Performance for Demanding Edge Workloads