For reasons of practical handling as well as optimization of the processes of development and implementation, it is desirable to realize realtime models of sound emitting physical processes in a modular fashion that reflects an intuitively understandable structure of the underlying scenario. At the same time, in discrete– time algorithms based on physical descriptions, the occurance of non–computable instantaneous feedback loops has to be avoided. The latter obstacle prohibits the naive cross-connection of input– output signal processing blocks. The following paper presents an approach to gain modularity in the implementation of physicsbased models, while preventing non–computable loops, that can be applied to a wide class of systems. The strategy has been realized pratically in the development of realtime sound models in the course of the Sounding Object [1] European research project.