@dynamo
I am not quite sure what you are trying to say but if you applied a RLE algorithm to a truly random stream of data then your "compressed" data would most likely be larger than the original stream.
Treating the data as a stream of bytes remember in RLE you either have the RLE every byte or you have to mark the the sequences that are RLE. In the first case given any byte to encode then 255/256 (or 99.6%) of the time the byte that follows is different and I have to encode 1 byte into 2. With other 0.4% of the time if the byte following the second byte is different then 2 bytes is encoded in 2 bytes so there is no compression. It is only if I have 3 bytes (or more) the same that I actually get compression. There is only a 0.0015% chance of this.
Similarly if I choose to mark the compressed runs of data using some byte value M then a minimum RLE code is 3 bytes (M, 1 byte containing the count, 1 byte containing the value). That means that to compress the data I need a run of at least 4 bytes which only occurs 0.000006% of the time. And offseting that every time M appears in the original stream without M following it (about 0.38% probability) I have to find a way to mark it, normally by repeating it twice in the output stream.
The point is that RLE only really works as a compression algorithm when you know that mainly what you are dealing with is long runs of data that is the same so that you are guaranteed good compression; random data does not meet this caveat you are equally likely to get short runs of data as long ones.
This basic point is true of every compression algorithm. Any given algorithm is tuned to compress data of a given pattern (or patterns). Data not of that pattern will at best not compress under the algorithm and in at least some worst cases the algorithm will cause the data to be expanded.
That is why if you have data you wish to compress it is important to choose the right algorithm for the patterns in the data.
Random binary data by its definition has no pattern and therefore you are unable to choose a algorithm that will compress it. Any algorithm you chose will probably compress some parts of the stream but it will also certainly expand other parts of it.