Service offloading poses interesting challenges in current and next-generation networks. The classical network optimisation algorithms are still painstakingly tune heuristics to get a sufficient solution. Classical approaches use data as input in order to output near-optimal solutions. These techniques are show exponential computational time and deal only with small network scale. Therefore, we are motivated by replacing this tedious process with recent learning techniques to learn the behavior of the classical optimization algorithms while enhancing both the quality of service and satisfying the resources requirements of next-generation applications. Deep Reinforcement Learning (DRL) and Machine Learning (ML) can improve service offloading and network caching. An optimal service offloading in virtual mobile edge computing (SO-VMEC) use case algorithm is proposed using Integer Linear Programming (ILP). Moreover, a service offloading protocol is presented to support the use case. We leverage Software Defined Networking (SDN) and Network Function Virtualization (NFV) concepts to control and virtualize network components. Then, a DRL-based offloading is proposed to deal with dense Internet of Things (IoT) networks. Extensive evaluations and comparison to state of the art techniques are carried out. Results show the efficiency of the proposed algorithms in terms of service offloading, resource utilization, and networking.