Less is More: Optimizing Function Calling for LLM Execution on Edge Devices