Empirical Guidelines for Deploying LLMs onto Resource-constrained Edge Devices