Run time power management has been a major problem in edge devices. Similar problem in High-Performance Computing is resolved using approaches like Q-learning. The bottleneck of implementing Q-learning on edge compute platforms is the large Q-table size and the compute load of algorithm. As part of our work, we propose a lightweight Q-learning methodology with memory management and work-load management policy. Proposed Run Time Manager eventually helps to bring down the power. Our proposed methodology has been implemented on Jetson Tx2 board which brought run-time power savings between 5%-23% and Q-table size decreased by 60%