Edge ML faces resource and energy constraints, requiring optimized split computing, which partitions inference be- tween edge and cloud. We propose a two-phase framework combining offline optimization and dynamic scheduling. It jointly configures split points and hardware settings to bal- ance energy and latency.