The European Commission project SoftEnable extends the concept of rigid body caging to soft, deformable objects, integrating both extrinsic and intrinsic constraints to create robust manipulation primitives capable of handling perturbations effectively. The project is a part of SoftEnable and focuses on designing tools or robot end-effectors using iterative bilevel optimization. A Bayesian optimization framework proposes candidate designs in the outer loop, and the Berzelius computing resources are used to train an image-based RL agent to learn manipulation policies to evaluate these designs. The reinforcement learning agent processes RGB images of manipulated rigid/deformable objects as observations and outputs corresponding tool/gripper actions. Our novel approach involves designing for robust manipulation (e.g., pushing, scooping) of deformable objects, using manipulation robustness metrics as rewards in the training process.