Illicit drugs are often mixtures containing cutting agents and other adulterants. The adulterants can cause harm on their own or when consumed in conjunction with the illicit drug. Drug checking services mitigate the risks associated with illicit drugs by providing substance composition and other harm prevention information. A widely embraced delivery model for drug checking involves offering the service at events and festivals. To help reduce drug-related harm at festivals, it is beneficial to have substance identification methods that are timely, safe, portable, and easy to use. Near-infrared spectroscopy (NIRS) is a well-known method for identifying and quantifying a range of substances, including the analysis of illicit drug mixtures, particularly when using high-end bench-top instruments. Providing mixture analysis by a portable NIRS solution is advantageous, however, there are still challenges due to device limitations. Consequently, a variety of methods have been utilised on portable NIRS data of various mixtures. Nevertheless, evolutionary computation methods, known to be robust for solving complex combinatorial and optimisation problems, are not as frequently reported in NIRS. This paper proposes a genetic programming-based approach to develop models for analysing illicit drug mixtures using portable NIRS. The experiment results indicate that the proposed approach can effectively provide a model for the analysis of mixtures that provides an accurate identification of components within a drug sample that is comparable to traditional linear-based models.