The requirement of colour reproduction has long been a priority driving the development of new colour imaging systems that maximise human perceptual plausibility. This thesis explores machine learning algorithms for colour processing to assist both content production and distribution. First, this research studies colourisation technologies with practical use cases in restoration and processing of archived content. The research targets practical deployable solutions, developing a cost-effective pipeline which integrates the activity of the producer into the processing workflow. In particular, a fully automatic image colourisation paradigm using Conditional GANs is proposed to improve content generalisation and colourfulness of existing baselines. Moreover, a more conservative solution is considered by providing references to guide the system towards more accurate colour predictions. A fast-end-to-end architecture is proposed to improve existing exemplar-based image colourisation methods while decreasing the complexity and runtime. Finally, the proposed image-based methods are integrated into a video colourisation pipeline. A general framework is proposed to reduce the generation of temporal flickering or propagation of errors when such methods are applied frame-to-frame. The proposed model is jointly trained to stabilise the input video and to cluster their frames with the aim of learning scene-specific modes. Second, this research explored colour processing technologies for content distribution with the aim to effectively deliver the processed content to the broad audience. In particular, video compression is tackled by introducing a novel methodology for chroma intra prediction based on attention models. Although the proposed architecture helped to gain control over the reference samples and better understand the prediction process, the complexity of the underlying neural network significantly increased the encoding and decoding time. Therefore, aiming at efficient deployment within the latest video coding standards, this work also focused on the simplification of the proposed architecture to obtain a more compact and explainable model.