A large eddy simulation study is undertaken with the objective of improving carbon monoxide (CO) estimations compared to measurements of enclosed turbulent flames in laboratory scale burners. The sub-grid combustion is modelled using a presumed joint probability density function with tabulated chemistry based on unstrained flamelets. The thermochemical conditions are mapped using the mixture fraction and reaction progress variable. A second progress variable is used to represent the post-flame oxidation of CO. The CO mass fraction is transported, and the CO production and consumption rates are stored in separate look-up tables using each progress variable. The two-progress variable model is assessed using enclosed flames that are stabilised behind a bluff body and in a gas turbine model combustor with swirling flows. The performance of the two-progress variable model is assessed by using standard practice approaches to compute the CO mass fraction with a single progress variable, which include obtaining the CO mass fraction from a look-up table or from its transport equation. Transporting the CO mass fraction with a single progress variable gives significant overestimations in the burnt regions of the bluff body stabilised flame. The two-progress variable gives significant improvements in the burnt regions, since the disparate time scales of CO production and consumption are captured by this model. The look-up approach also significantly overestimates the CO mass fraction in the flame stabilisation region in both configurations, where convection and diffusion processes are dominant. Therefore, the CO mass fraction needs to be transported with two progress variables to capture the intricate convection-diffusion-reaction balance in the flame stabilisation region and within the flame brush. Furthermore, the two-progress variable model gives improvements in the post-flame oxidation regions towards the chamber exit in the gas turbine model combustor, whereas the look-up approach gives significant overestimates.