Composite structures as in UML are a way to ease the development of complex applications. Composite classes contain sub-components that are instantiated, interconnected and configured along with the composite. Composites may also contain operations and further attributes. Their deployment on distributed platforms is not trivial, since their sub-components might be allocated to different computing nodes. In this case, the deployment implies a split of the composite. In this paper, we will motivate why composites need to be allocated to different nodes in some cases by examining the particular case of interaction components. We will also discuss several options to achieve the separation and their advantages and disadvantages including modeling restrictions for the classes.