Many problems in astrophysics are multi-physics and multi-scale in nature, involving the evaluation of nuclear equation of state (EOS) and opacity for radiation transport, which are often computationally expensive. Our study explores using machine learning techniques in one of the underlying computational building blocks of the general astrophysics simulation system (GenASiS). Specifically, we integrate a pre-trained neural network (NN) model into GenASiS, replacing a kernel process. Upon pre-training our experimental model with data generated by the Riemann problem, we tested it on a complex problem like the Kelvin-Helmholtz instability (KHI) problem. We observed the model’s ability to accurately reproduce solutions without compromising runtime efficiency. This success can be attributed to shared aspects between the two problems, including their similar nuclear EOS and fundamental physical principles. The study is therefore also an illustration that cross-over machine learning approaches can benefit problems with similar foundations.