The world is moving towards a new generation of Internet, the Internet of Things (IoT). This technological jump is improving the way we live by creating a bridge between the physical and the virtual world. Researchers are curious about the field from different perspectives especially managing of the complexity and dynamicity of these systems. This study is about the necessity of having a mechanism that enables IoT systems to perform without breaks or shutdowns regardless of context changes. The solution that we propose consists of a contextual dynamic reconfiguration process to be implemented by a reflexive multilayered architecture. This process is based on the autonomic computing MAPE/K loop. We also integrated the use of evolution styles to make reusable the reconfigurations applied on the architecture of the system. Validation of the proposed approach was made on an e-health scenario, which was simulated using Cisco Packet Tracer before performing real development.