Mission-critical system typically refers to a project or system for which the success is vital to the mission of the underlying organization. The failure or delayed completion of the tasks in mission-critical systems may cause severe financial loss, even human casualties. For example, failure of an accurate and timely forecast of Hurricane Rita in September 2005 caused enormous financial loss and several deaths. As such, real-time guarantee and reliability have always been two key foci of mission-critical system design. Many factors affect real-time guarantee and reliability. From the software design perspective, which is the focus of this paper, three aspects are most important. The first of these is how to design a single application to effectively support real-time requirement and improve reliability, the second is how to integrate different applications in a cluster environment to guarantee real-time requirement and improve reliability, and the third is how to effectively coordinate distributed applications to support real-time requirements and improve reliability. Following these three aspects, this dissertation proposes and implements three novel methodologies: real-time component based single node application development, real-time workflow-based cluster application integration, and real-time distributed admission control. For ease of understanding, we introduce these three methodologies and implementations in three real-world mission-critical application systems: single node mission-critical system, cluster environment mission-critical system, and wide-area network mission-critical system. We study full-scale design and implementation of these mission-critical systems, more specifically: 1) For the single node system, we introduce a real-time component based application model, a novel design methodology, and based on the model and methodology, we implement a real-time component based Enterprise JavaBean (EJB) System. Through component based design, efficient resource management and scheduling, we show that our model and design methodology can effectively improve system reliability and guarantee real-time requirement. 2) For the system in a cluster environment, we introduce a new application model, a real-time workflow-based application integration methodology, and based on the model and methodology, we implement a data center management system for the Southeastern Universities Research Association (SURA) project. We show that our methodology can greatly simplify the design of such a system and make it easier to meet deadline requirements, while improving system reliability through the reuse of fully tested legacy models. 3) For the system in a wide area network, we narrow our focus to a representative VoIP system and introduce a general distributed real-time VoIP system model, a novel system design methodology, and an implementation. We show that with our new model and architectural design mechanism, we can provide effective real-time requirement for Voice over Internet Protocol (VoIP).