This dissertation is concerned with the design and implementation of infrastructural systems for resilient Machine-to-Machine (M2M) communication in distributed Cyber-Physical Production Systems (CPPS). For this purpose, a number of technologies are selected and applied in congruence with the principles of Smart Manufacturing. Thus, this dissertation investigates the use of Service Oriented Architectures (SOA), M2M communication middleware systems, overlay networking solutions, and other technologies to improve the agility, resilience, and interoperability of manufacturing infrastructure. First, the concept of SOAs and how they may be applied in current enterprises to achieve flexibility, agility and interoperability is addressed. As such, the technical state of current industrial enterprises and the characteristics of the Service Oriented (SO) approach are detailed to highlight the competitive advantage possible through service orientation. A review of preliminary SO Reference Architectures (RA) delivered by major European Union (EU) research projects is conducted to determine their features and possible shortcomings. Realisations of the architectures are also discussed to underline their choices in technologies and their delivered technical innovations. Based on the findings of the review, the SO Open Platform Communications Unified Architecture (OPC UA) is selected as the base technology for the envisioned system. Following a bottom-up approach to system development, this dissertation proceeds to investigate the rigid networking infrastructure in manufacturing enterprises. It evaluates the possibility of using Peer-to-Peer (P2P) networking technologies to create a cohesive, fault-tolerant network of components for the non-real time management, replication, storage, and sharing of plant data and service components. A cooperative P2P overlay network is proposed as the most applicable architecture for manufacturing systems. While the overlay network proposed may be used as a transport layer for OPC UA, it is in fact developed as a middleware-agnostic protocol, thus affording it wider applicability in the domain. The design is also evaluated through a prototypical implementation that demonstrates the viability of the approach. Finally, the OPC UA specifications are reviewed to highlight possible items for enhancement. Specifically, two proposals are evaluated. The first assesses the practicality of employing a dedicated service for the distributed coordination of redundant OPC UA servers. The second applies a queueing service to shield resource-constrained OPC UA servers from high rates of concurrent asynchronous service calls that may lead to resource exhaustion. Both proposals are evaluated programmatically and the code is open sourced. Results demonstrate the feasibility of their respective designs.