Using simulation to better understand the mechanical response of parts and assemblies can be an integral part of many product design processes, as well as being potentially instrumental in tuning assembly line processes. However, solving complex problems with Finite Element Modelling (FEM) can be computationally expensive, especially for a Design of Experiment (DoE) that typically requires many individual simulations. In this study a medical device was used to demonstrate the expensive DoE process, as well as how to make better use of DoE databases using surrogate modelling. Traditionally, Reduced Order Models (ROMs) have been constructed to reduce the computational cost associated with simulations. ROMs can, however, be insufficient for capturing the complex nonlinearities of many problems. In recent years, machine learning has emerged as an attractive alternative to overcome the shortcomings of ROMs while also reducing the computational effort. More specifically, Gaussian Process Regression (GPR) has become a popular algorithm for machine learning based surrogate models due to its ability to output a probability distribution rather than a single point estimate (typical of a neural network for example). In this work, we propose a surrogate model of a simple autoinjector medical device assembly and disassembly, and demonstrate that this model is not only capable of predicting the mechanical response of the autoinjector accurately, but can also steer a designer of the confidence in the prediction. Autoinjector medical devices are complex tightly packaged assemblies, and it is common to experience significant variation in assembly forces for key components and subassemblies due to component variability in injection moulding manufacturing processes. The training set used in this study was based on a series of FEM simulations computed using Ansys LS-Dyna, in which geometry representing an autoinjector syringe holder was assembled into an outer casing (typical of a common autoinjector assembly step). A Monte Carlo approach was used create the database, which contained 500 individual randomised assemblies based on typical injection moulding process variability. Gaussian process regression is a probabilistic method based on multivariate Gaussian distributions. GPR learns the function mapping between the inputs and outputs in a supervised manner. The prediction of a GPR is a distribution described by a mean and a standard deviation which can be interpreted as the predicted value and the uncertainty associated with the prediction, respectively. The GPR building process developed for this study is presented. Force trace data from key components outputted from the simulation were discretised evenly into suitable displacement intervals and the GPR approach used to predict the mechanical response (force) of the autoinjector components during an assembly/disassembly step. The output was a GPR based surrogate model which could be rapidly trained and queried to aid assembly line decision making, qualifying a new source of part supply for successful assembly, and providing near instant steers on potential assembly line yields based on known process force limits. GPR offers an attractive machine learning approach for surrogate modelling that can significantly speed up the product design process. The surrogate model proposed in this work can be leveraged at multiple stages of the design loop to carry out optimisation and/or uncertainty quantification, and later on to de-risk during the manufacturing process.