Soft real-time systems are normally defined as those which are having performance constraints on several time measures. Usually, design methods for large real-time systems do not provide any help in performance analysis and estimation of the system which is being designed, for this reason the main proposal of the paper is how can they be extended to cope with this subject. This paper presents several approximate analytical models of performance software elements, called multiclass channels and pools. These elements are used in the construction of large real-time software systems to complement automatically the system design, in order to obtain the desired average estimates for the soft real-time constraints. Channels transfer data between tasks without synchronisation in a selective manner while Pools do it in a non-selective way.