Far below (in section "original 3 steps"), you see the following:
1. a temp table is created
2. some data is inserted into this table
3. some of the inserted data is removed based on a join with the same
table that the original select was made from

In my opinion, there is no way that the join could produce more rows
than were originally retrieved from viewD. Hence,  we could get rid of
the DELETE step by simply changing the query to be:

INSERT INTO #details ( rec_id, orig_corr, bene_corr )
SELECT rec_id, 0, 0
FROM   viewD
WHERE  SOURCE_SYS NOT IN ( 'G', 'K' )
AND    MONTH( VALUE_DATE_A8 ) = MONTH( @date )
AND    YEAR( VALUE_DATE_A8 )  = YEAR( @date )
AND  INMESS NOT LIKE '2__'    ---- the added line

===== original 3 steps (mentioned above) =====

CREATE TABLE #details (
   rec_id  UNIQUEIDENTIFIER PRIMARY KEY NOT NULL,
   orig  VARCHAR(35) NULL,
   bene  VARCHAR(35) NULL,
   orig_corr   TINYINT     NULL,
   bene_corr   TINYINT     NULL
)

INSERT INTO #details ( rec_id, orig_corr, bene_corr )
SELECT rec_id, 0, 0
FROM   viewD
WHERE  SOURCE_SYS NOT IN ( 'G', 'K' )
AND    MONTH( VALUE_DATE_A8 ) = MONTH( @date )
AND    YEAR( VALUE_DATE_A8 )  = YEAR( @date )

DELETE d
FROM   #details d
JOIN   viewD v ON ( d.rec_id = v.rec_id )
WHERE  INMESS  LIKE '2__'