Corno,
First of all, I think I speak for a few people when I ask why you want this approach.
Second, I have started to work on solving your problem. Without an understanding of the reason for this, I am having trouble with the implementation.
It is quite easy to apply the concept, as I think I understand it in LabVIEW. Yes, dataflow can be ignored, and even reversed in LabVIEW. It is not how the language was designed, but it can be done. The implementation I came up with was to use VI server or possibly shift registers to handle your parameters. A flag can be set (or a null value used) to determine if the parameters have been calculated when they are needed.
The problem that you have with this is that you don't know what the memory allocation or ti
ming will be. You can know the maximum amount of memory that will be needed, but timing of the program execution will be dependent on whether or not parameters are 'filled' and when.
I guess if those items aren't critical, then you won't have a problem.
The other idea here is that your processor or operating system may not efficiently support such a notion. For example, I have heard that IBM created a supercomputer long ago that specialized in recursion. Hardly a common need, but this computer was really good at it.
If you could provide some more detail as to why you need this, I am sure that someone here can either offer a solution, or an alternative.
For now, give the VI server a try.
Good luck