Researchers in cognitive robotics have developed different architectures to help social robots make decisions and express emotions autonomously in order to replace the pre-programmed and remote-controlled techniques. However, most of the current developments are ad-hoc solutions with no possibility to be utilized in multiple therapeutic situations. In this paper, we propose the design of a social behavior architecture, which aims at helping the social robot to sustain the user’s engagement and motivation, and to achieve the goal of interaction in different scenarios. Representations of behaviors are kept at abstract level and mapped with the robot’s morphology afterward. This approach ensures the architecture to be applicable to a wide range of social robots.