The prediction of temperatures in power semiconductor modules, such as insulated-gate bipolar transistors (IGBTs) is critical to ensure adequate lifetime modeling of the devices. A temperature of particular interest is that of the semiconductor junction, which is used to assess the lift-off of wire bonds. For many applications featuring dynamic loads, the junction temperature needs to be simulated for so-called mission profiles of significant duration. To limit the computational expense, the simulations are based on thermal impedances from junction to ambient, which may be obtained from numerical 3-d simulations. Even these 3-d simulations can be computationally expensive. In power-electronic systems, often, large heat sinks are used with a multitude of mounted IGBT modules, interacting thermally. In such cases, the detailed 3-d models become large and the transient simulations are not feasible. In the present work, a method is proposed that allows us to significantly reduce the 3-d model size. To this end, the ideas of compact or boundary-condition-independent models are used. The presented method has the advantage that, unlike in model-order reduction, the system matrices of the 3-d model are not needed. This makes the method applicable to commercial simulation software like ANSYS Icepak™, that does not give access to the system matrices. The method is implemented via MATLAB™ scripts that automatically generate 3-d ANSYS Icepak™ models of IGBT modules on a heat sink. An example case of two IGBT modules mounted on an air-cooled heat sink is presented, and the method is shown to yield good accuracy (thermal-impedance errors below 8% and thermal-resistance errors close to zero), while reducing the model's mesh size by the factor of 14. Further error reduction is expected to be possible by adapting the model parameters. This can be subject to future work.