M2M systems need to connect thousands of various fix and mobile machines that are widely distributed and evolve frequently according to their profile and context changes. The increasing complexity of managing current distributed systems needs new solutions. Designing autonomic systems, which are self-managing, and context aware is a solution and also a challenge. This paper proposes the FRAMESELF framework, a generic autonomic architecture based on decision models. The proposed solution capabilities are used for the self-deployment and self-configuration of machine-to-machine (M2M) systems. Components diagrams are illustrated to describe FRAMESELF modules and to show how they are connected together. FRAMESELF implements Devices Profile for Web Services (DPWS) protocol to describe and to discover managed resources. It implements multi models representation based on ontologies and graphs to describe the M2M concepts and relationships on a multi-level knowledge base. Two communication patterns modules based on service-oriented and event-driven communications are dynamically selected and configured in deployment plans. Finally, a smart metering scenario is experimented to validate this approach and to calculate the overload that FRAMESELF generates facing scalability. INTRODUCTION Over the last years, Internet of Things (IoT) has evolved at an exceptional speed. Things can both correspond to physical things (sensors, actuators, smartphones, machines, etc.) or immaterial ones (applications, multimedia content providers, directories, etc.). Such environments consist of a high amount of heterogeneous entities having mutual interactions and delivering high-level services that could be discovered, monitored, composed, and executed. One of the main concepts behind IoT is Machine-to-Machine (M2M) communication. M2M provides seamless integration of the heterogeneous participating machine domains overcoming the interoperability issues raised between them. The growing number of interconnected machines and the diversity of communication technologies and protocols as well as the increasing volume of exchanged data mirrored the increase of complexity in M2M systems. Deployment, configuration and maintenance of M2M platforms are costly in term of time and money and require a permanent presence of high skilled administrator. For this purpose, it is paramount to provide efficient M2M systems with capability of self-management to increase the autonomic potential of M2M systems. On the first hand, autonomic computing paradigm was introduced by IBM to deal with system complexity which is inspired by the human autonomic nervous system that handles complexity and uncertainties, and aims at realizing computing systems and applications capable of managing themselves with minimum human intervention. In theory, Autonomic computing encompasses four self-management capabilities including self-configuring, self-healing, self-optimizing and self-protecting. Up until the present time, available autonomic systems are restricted to specific problems and are not applied in M2M systems due to the vertical fragmentation of M2M application domains. On the other hand, ETSI defined