Advances in medical imaging have enabled detailed digitisation and representation of human anatomy, but challenges remain when modelling complex structures. Statistical models have been developed to capture variations in shape, pose, and intensity features of anatomical structures. However, these models often embed a single feature. This paper investigates how a novel dynamic multi-feature-class Gaussian process modelling framework (DMFC-GPM) designed to learn shape, pose, and intensity features in continuous domains, facilitates the modelling of complex anatomical structures. The work evaluates the framework’s ability to capture multi-feature variations within complex anatomy. Computed tomography image data was processed to build and validate a statistical shape, pose and intensity neutral-arched foot model (12 bones). Framework evaluation was done by validation of the model using specificity, and generality. Fitting the model globally to all objects resulted in specificity and generality reported as average root mean square (RMS) of 0.61 ± 0.11 mm and 1.02 ± 0.21 mm, and average Hausdorff distance (Hd) of 3.47 ± 0.98 mm and 7.56 ± 1.33 mm, respectively. Further validation of the model marginalised to the talus bone resulted in specificity and generality of 0.81 ± 0.25 mm and 1.20 ± 0.48 mm average RMS and 3.24 ± 1.10 mm and 6.50 ± 2.34 mm average Hd, respectively. The talus model variations were consistent with literature. Thus, the novel DMFC-GPM framework can model complex anatomies such as the foot arch.