Dynamic fuel cycle simulation codes model evolving nuclear fuel cycles, and calculate nuclides inventories and material flows in each unit of the cycle. In the nuclear fuel cycle simulation code CLASS (Core Library for Advanced Scenario Simulation), a Fuel Loading Model (FLM) builds a fresh fuel fulfilling the reactor criticality requirement, depending on the available fissile material. Then, a mean cross-sections predictor calculates the mean cross-sections required to perform the fuel depletion in a short calculation time. This work focuses on the elaboration of these models in the case of a PWR-MOXEUS fuel (MOX on Enriched Uranium Support), which allows plutonium mono-recycling and multi-recycling in PWR. These models are built using neural networks. These predictors are trained on a databank composed of 1000 PWR infinite assembly depletion calculations performed using the software MURE (MCNP Utility for Reactor Evolution) based on the transport code MCNP (Monte-Carlo N Particle). Several databanks are tested and the performance of the resulting predictors are compared. The FLM predicts the plutonium content, and potentially the uranium enrichment, required in the fresh fuel. This model is based on a calculation of the infinite multiplication factor performed with an accuracy close to MCNP statistical error. Mean cross-sections prediction allows a deviation lower than 5% on main plutonium isotopes at 75 GWd/t compared to the fuel depletion reference calculation. PWR MOXEUS models are also tested on a balancing scenario. The complex evolution of MOXEUS fresh fuel isotopic composition during the scenario is highlighted. Furthermore, equilibrium fresh fuel isotopic vectors are compared to another study on an equilibrium MOXEUS multi-recycling strategy calculation, showing a good general agreement.