Researchers are increasingly interested in early life exposure to complex mixtures, and their effects on childhood neurodevelopment. These studies typically use models that assume exposure mixtures affect each child's development to the same degree. In reality, however, there may be latent, or underlying, population subgroups of children with distinct neurodevelopmental trajectories over time. Studies have not addressed if children in latent classes with distinct neurodevelopmental trajectories - particularly those in the delayed latent class - are differentially vulnerable to complex mixtures. They also have not studied if interactions among the mixture components differ across latent classes. Hence, we propose a two-stage statistical model and conduct a secondary data analysis of PROGRESS, a prospective cohort study. Our model first uses growth mixture modeling to identify latent classes of infants with distinct patterns of neurodevelopment. Second, we separately study each latent class, using a flexible approach called Bayesian varying coefficient kernel machine regression, to model how metal mixtures, and their interactions, are associated with neurodevelopmental trajectories.