I think where you might have a misunderstanding is thinking of the miners and the clients as separate, but it absolutely requires both mechanisms for the bitcoin network to function.  And as I pointed out, they where one in the same software originally.  It is technically feasible to create and sign a transaction without doing any mining.  It is also feasible to send it "out of network" by printing it on a piece of paper and mailing it to someone for example.  At any point in time that transaction could be transmitted to the network and then be validated as long as the original bitcoins had not moved. But until the transaction is actually incorporated into the block chain it has no meaning to the network at all.  

In terms of implementation there is a pool of transactions that each miner keeps that it hears about from other nodes.  At each miners desecration (based on fees or other criteria) the miner can choose to incorporate any or all of the transactions it has in it's pool of unconfirmed transactions into the next block it attempts to solve.  These pools are more or less unique to each miner, there isn't a "big pool in the sky" of unconfirmed transitions.  Once a miner confirms a transaction in it's own pool, or receives a confirmed block that has that transaction the minor removes the transaction from it's pool of unconfirmed transactions, and continues on with the next set of unconfirmed transactions.

Is that perhaps more helpful?  Sorry if I came across a snippy, but you can't ask technical questions and then immediately restrain people from giving you the correct answer.