We consider a system where multiple users are connected to a small cell base station enhanced with computational capabilities. Instead of doing the computation locally at the handset, the users offload the computation of full applications or pieces of code to the small cell base station. In this scenario, this paper provides a strategy to allocate the uplink, downlink, and remote computational resources. The goal is to improve the quality of experience of the users, while achieving energy savings with respect to the case in which the applications run locally at the mobile terminals. More specifically, we focus on minimizing a cost function that depends on the latencies experienced by the users and provide an algorithm to minimize the latency experienced by the worst case user, under a target energy saving constraint per user.