Computing architectures have changed significantly since their inception. Adapting software to such changes can be a serious challenge. The DEM platform Aspherix is based on LAMMPS, a molecular dynamics solver optimized for computing on CPUs with distributed memory using the MPI library. While such architectures remained at the forefront of high-performance computing for an extended period, they have been largely superseded by heterogeneous CPU-GPU clusters. Moreover, desktop computers often feature GPUs capable of general-purpose computing, e.g., using the CUDA library. Predicting future compute architectures is destined to fail, and thus porting a large code base to a specific platform is not future-proof. To circumvent this issue, a performance-portable programming model is used. It allows for rewriting the Aspherix software with little regard to the underlying hardware. Nevertheless, adapting the algorithms from single to multi-threaded requires a significant effort, and several needed to be reimplemented. This work presents some of the necessary strategies on how to approach such changes. The novel implementation automatically chooses the optimal variant for the available hardware and can perform computations using CUDA or OpenMP libraries while retaining the capability for multi- node compute clusters. The currently available features exhibit significant performance increases compared to the traditional software variant on specialized hardware while keeping the overhead minimal when run on the CPU. We will show several simulations alongside an analysis of their performance and outline further work to be done in porting the code base.