Aviation is an important contributor to worldwide greenhouse gas emissions. This thesis aims to help reduce NOx emissions by proposing a gradient-based optimisation framework that allows to implement NOx emission constraints from the preliminary design phase. Simple chemical reactor networks built on Cantera are used to predict NOx emissions of Jet A and hydrogen combustors. The models are used to train Kriging surrogate models with the Surrogate Modelling Toolbox, which provides analytical gradient information. The gradient-enabled surrogate models are integrated into pyCycle, a modular, gradient-based engine cycle analysis tool built on OpenMDAO. The models are used to implement NOx constraints during the optimisation of conventional turbofans, and turbofans with water recirculation. This work shows that it is possible to implement NOx emissions constraints during the preliminary design phase, and that a 40% reduction in NOx can be achieved with a maximum fuel penalty of 1.2% across the analysed cases.