Spinal cord injury (SCI) disrupts communication within central nervous system and lead to range of neurological disorders including paralysis. Current rehabilitation strategies to restore locomotion are poorly effective in people with severe SCI. Epidural electrical stimulation (EES) showed promising results in animal models and humans to improve recovery. However, EES protocols have remained empirical, which has limited the efficacy of this paradigm. In this thesis, I report my contribution to the development of EES related software and hardware that were driven by the under- standing of the mechanisms underlying EES. These neurotechnologies showed a remarkable efficacy to reestablish locomotion in animal models of leg paralysis. EES protocols have primarily been delivered with continuous stimulations applied over the middle of the spinal cord, independently of current limb positions. However the natural activity of the spinal cord does not correspond to this limited monotone modulation. In the first part of this thesis, we developed spatiotemporal stimulation protocols that seek to reproduce the natural activity of the spinal cord during walking. Using computational models, we identified optimal electrode locations that target individual posterior roots. These simulations steered the design of spatially selective spinal implants. Control software triggered stimulation trains in real-time to engage extensor and flexor synergies according to limb positions. This spatiotemporal neuro- modulation therapy enhanced locomotor performances in rats with severe SCI. However, these spatiotemporal neuromodulation protocols were controlled via an external computer. The animal had no control over the occurrence of the stimulation. To remedy this limitation, we designed a brain-spine interface. We linked motor states decoded from leg motor cortex activity to spinal cord stimulation protocols to elicited the intended movements. We implemented this brain-spine interface in a nonhuman primate models of transient unilateral leg paralysis. All the animals immediately regained movements of the paralyzed leg. There is often a mechanical mismatch between current stiff implants and the soft neural tissues of the host. The third part of this thesis is dedicated to the design and fabrication of soft neural implants that present the same elasticity as the dura mater. These implants maintain the ability to deliver electrochemical stimulation while being stretched. Such interfaces can be applied to multiple neuroprosthetic applications, including the restoration of locomotion in paralyzed rats. In the last part of this thesis, I addressed the current limitations of spinal implant specificity. We found that EES activates the large afferent fibers within the dorsal rots and developed novel implants that take advantage of the known trajectories of the dorsal roots to increase the specificity of EES. First, we built a precise anatomical computational model of the lumbosacral roots. We then ran simulations that identi- fied multipolar electrode configurations capable of steering the current specifcally. These results drove the fabrication of soft neural implants that we inserted chronically in rat models of severe SCI. Experiments showed that multipolar EES increases the specificity of the stimulation compared to conventional protocols. All the concepts developed and validated in my thesis are already applied or are steering applications in humans with SCI.