In Brain-Computer Interfacing (BCI), brain activity is recorded and translated into user-intended actions. Several studies have successfully employed motor-BCIs to replace the function of a lost or impaired limb by circumventing disconnected neural pathways. Electrocorticography (ECoG), a partially invasive recording technique, offers new perspectives for long-term recording as it avoids fibrous scar tissue formation of electrodes implanted in the cortical tissue. Despite advances in individual finger decoding based on ECoG, including ours, demonstrations of more realistic scenarios such as repeated finger movements (cf. typing) and coordinated finger actions (grasping and holding a cup) are still lacking. Our main objective is to develop an ECoG-BCI for controlling self-paced individual- and coordinated finger movements with a hand exoskeleton, a wearable motorized framework. As current decoders, operating under the assumption of a linear relationship between finger actions and ECoG activity, do not perform well in these scenarios, a new nonlinear decoder is proposed. Our second objective is to address subject training. We propose a multi-step training strategy that will assist the user in gradually acquiring finger dexterity from motor imagery, as it is the latter on which paralyzed individuals can rely. We expect the hand exoskeleton to yield beneficial training- and usage effects, as it allows for tactile feedback of the intended actions.