Analog memristors are key building blocks to accelerate inference and training workloads of modern neural networks. This work provides a physical understanding and sets the design rules for novel filamentary Conductive Metal Oxide (CMO)-HfO2 Redox-based Resistive Switching Random Access Memory (ReRAM) devices. These Metal/CMO/Insulator/Metal structures show superior characteristics in terms of reduced switching stochasticity, higher number of non-volatile states and reproducibility upon switching with respect to conventional Metal/Insulator/Metal technology. The experimental data are described using a physics-based model in COMSOL Multiphysics 5.2 software. The simulations reveal the presence of a spreading resistance in the CMO acting as a current bottleneck in the Low Resistive State (LRS) while the oxidation of the CMO in a dome shape determines the High Resistive State (HRS). In addition, the role of the thickness and the electrical conductivity of the CMO, as well as the radius of the conductive filament, are explored for analog CMO-based ReRAM devices.