Production Planning Management and Control of a production system subject to random events are challenging problems. A multi-layer hierarchical approach to tactical and aggregate production planning problems is proposed, wherein the architecture is strongly based on the specific physical system, applicable controls and the complexity of the decision making problem at hand. We address the design and operation of such Hierarchical production Management Systems (HPMS). Regarding the design aspect, we start by developing schemes for product, machine and temporal aggregation; consistency and controllability issues in the hierarchy have been addressed in the aggregation/disaggregation schemes. These three aggregation schemes for model reduction have been developed and incorporated to the time-scale decomposition of activities, in order to provide a solid theoretical foundation of the architecture. We then proceed to a systematic stepwise design approach for the construction of the hierarchy. It provides the appropriate number of layers and an associated Model as well as Decision Making Problem (DMP) at each level. A model is defined by entities, attributes, links and domains, while a DMP is defined by a set of possible controls (decisions), constraints, and optimality criteria to be optimized over a planning horizon. The operation of the hierarchy consists of a topdown computation of controls, which calls for the resolution of an optimization problem at each level of the hierarchy. The solution of any problem in sequence determines some parameters in the subsequent problem. We detail the mechanism for top-down constraint propagation, which is important in ensuring consistency of criteria and feasibility. The execution then involves the bottom-up feedbacks, and a revision in the plan is carried out if necessary. In particular, the rolling horizon mechanism, and the reaction of the hierarchy to random events has been detailed. A generic job-shop example has been employed to present the design and operation of the HPMS. It is hoped that this methodology can be applied to other types of large-scale complex decision making problems.