The emotion recognition field is a young but maturing field, for which needs for designing models begin to emerge.We consider emotion recognition to be a multimodal interaction with the machine. From this point of view, we designed a component-based architecture model for emotionnaly-wise interactive systems : the emotion branch. In this paper, we immerse our architecture model in the field of multimodal interaction. We redefine a modality within the frame of passive emotion recognition and show that doing so enables us to benefit from concepts from multimodal interaction (CARE properties and software specifications) and to highlight unexplored design spaces.