Using belief-propagation based algorithms like Max-Sum to solve distributed constraint optimization problems (DCOPs) requires deploying the factor graph elements on which the distributed solution operates. In some utility-based multi-agent settings, this deployment is straightforward. However, when the problem gains in complexity by adding other interaction constraints (like n-ary costs or dependencies), the question of deploying these shared factors arises. Here, we address this problem in the particular case of smart environment configuration (SECP), where several devices (e.g. smart light bulbs) have to coordinate as to reach an optimal configuration (e.g. find the most energy preserving configuration), under some n-ary constraints (e.g. physical models and user preferences). This factor graph deployment problem (FGDP) can be mapped to an optimization problem, then solvable in a centralized manner. But, when dealing with the dynamics of the environment (e.g. new sensed data which activates some rules, adding new devices, etc.) we cannot afford restarting the system or relying on a centralized solver. Thus, the system has to achieve on-line and local deployment adaptations. In this paper, we present some solutions and experiment them on a simulated smart home environment.