Models with loosely coupled subsystems and network models are often characterised by subsystems of different dynamics. Therefore, mulirate methods have been brought into the EOOLT community over the last couple of years. These methods split the system into a fast and a slow part and integrate them using different step sizes to obtain faster simulations. Besides the numerical properties of multirate methods, an important challenge is to divide the system into a manageable number of subpartitions which can be assigned to a suitable step size. This paper presents a promising approach towards system partitioning for multirate integration that is based on a systems task graph. The proposed algorithms and the guard mechanism aim to reduce the overhead of the multirate method which is introduced by deciding between active and latent variables at simulation execution time.